How your twin works

This page is the plain answer to "what is actually going on". It is here because a regulator, an App Store reviewer and a curious person all ask the same question, and the honest answer is short.

Your twin is software

It is a large language model from a third-party AI provider, listed on our recipients page, Who handles your information, and running in the United States — given a description of you and a set of rules, producing text. It is not a person, it is not conscious, and it is not you. Everywhere it speaks in the app there is a label above it saying so.

What it is made of

One interview, held once, with an AI host. From that we extract a structured profile:

• your first name, an age band, your city and neighbourhood, the languages you speak;
• your interests, with how deeply you are into each;
• what you are looking for;
• when you are usually free;
• boundaries — things you have told it never to bring up.

You see the whole profile before it is ever used, on a review screen, and nothing runs until you approve it. You can edit or delete any part of it at any time.

Every fact carries a flag: shareable or private. Personal facts are private by default. A private fact shapes how your twin behaves but is never sent to another person's twin, and there is a separate check on every sentence your twin produces that looks for private material having leaked into it — both word-for-word and paraphrased — and holds the sentence back if it finds any.

What happens overnight

1. Candidates: other approved twins in your neighbourhood cluster, excluding anyone either of you has blocked and anyone you have already been introduced to. 2. A model scores each pair on how likely they are to actually get on. 3. The best pairs have a real conversation — your twin and theirs, taking turns, using only the shareable parts of each profile. 4. The conversations are ranked and the best three become your morning.

You can read every word of the conversation your twin had. So can the other person. It is stored once and shown to both of you, because a conversation about two people belongs to both of them.

What it never does

• It never claims to be human. That rule is in the prompt, in the state machine, and in the database: a twin has no way to write a message into a chat with a person in it.
• It never talks to a person. Twins talk to twins.
• It never proposes meeting at a home.
• It never shares a fact you marked private, or anything inside a boundary.
• It never sees your phone number. That is not in the profile at all.

When a human reads it

Almost never. Nobody reads your interview or your chats as a matter of course. A person reads something when:

• you report someone, and the reported material is in that chat or that introduction;
• the injection screen holds a piece of text back and it needs a decision;
• you type a venue name that is not on the curated list, and it has to be checked before a plan is visible;
• you and the other person on a plan for two each say the other did not turn up. A moderator reads both answers and the note each of you attached, and decides; neither of you is shown the other's answer, whatever the decision;
• you have switched on "Helping us improve Churr" in Settings, in which case your twin's conversations can be read as examples when we test changes. Off unless you switch it on.

That is the list as of this version. If a new reason is added, this page changes first.

Where the thinking happens

The model belongs to a third-party AI provider and runs in the United States. Everything Churr *stores* is in Canada. The provider's commercial terms provide that inputs are not used to train its models. Push notifications go through a notification service, also in the United States. Both are named, with what each receives, on the recipients page.

Using your conversations to improve Churr

Off unless you switch it on, in Settings → Privacy. With it on, your twin's conversations can be used as examples when we test changes to the product — including in an automated test set that we run before shipping. With it off, none of your material is in that set, and switching it off later takes your material out.

It can be wrong

It can misremember what you said, overstate an interest, or describe you in a way you would not. The check-in loop exists to catch that, and Settings → What your twin shares is the direct fix. If your twin has said something about you that is wrong, that screen is the place to correct it, and the correction takes effect on the next run.

Turning it off

Settings → Pause my twin stops it going out overnight, keeps your chats open, and makes no new introductions. Settings → Delete my account removes it entirely, along with every conversation it had.

Questions: hello@churr.io.